Most people are using this week to mentally check out for the holidays.

Do that, and you’re gonna wake up in 2026 feeling 10 years behind in AI. 

While you are planning your time off, the AI wars just went nuclear with a 10 billion deal, a new best AI image model and a budget LLM that is already lapping its bigger, badder brother. 

(Yeah, don’t sleep on AI while you’re hitting snooze.) 

We’re keeping you up to speed in no time at all. 

Let’s get it. 

How'd we get from "Microsoft is OpenAI's ride-or-die" to "Amazon might drop ten billion on them" in like two months?

According to The Information, OpenAI is in confidential discussions with Amazon about a potential $10 billion investment and an agreement to use AWS's Trainium chips.

Here's the thing though. Microsoft has been OpenAI's primary backer since 2019. Over thirteen billion invested. But after OpenAI's October restructure? Microsoft no longer has exclusive compute rights.

Yuuuuup. The door is wide open now.

OpenAI recently valued at five hundred billion. Now there's reporting they're raising another round at eight hundred billion.

Compute is expensive. And OpenAI literally released a video saying they had to pause research because they don't have enough of it.

What it means: Google has really started to catch up on a lot of fronts.

This partnership could help OpenAI scale in 2026 and beyond.

When you need compute this badly, you make new friends fast.

According to The Wall Street Journal, Meta is preparing Mango, a new image and video model, and Avocado, their next-gen text model.

Both expected first half of 2026.

Holllllld up. Meta has been essentially silent for eight months on the LLM front. Spending tens of billions on acquisitions. Paying researchers hundreds of millions in short-term contracts.

And now? Fruit-named models.

But here's the catch. These new models may reportedly be proprietary. Previous Llama models were open source. You could use them for free.

That strategy might be dead.

What it means: The consumer AI race has shifted from text chat to trying to create viral visual content.

Google's Nano Banana Pro has gone absolutely viral. Meta is trying to close that gap.

Early 2026 will show if they come out cooking or swing and miss.

Anthropic just made a pretty interesting move. Their Claude for Chrome extension? No longer locked behind the $200/month Max plan.

Anyone on a paid Claude plan can now install it.

The plugin lets Claude fill forms, manage email and calendars, and complete multi-step workflows on your behalf. You know? The stuff you actually want AI to do.

Different strategy than the competition though. Perplexity and OpenAI are building agentic browsers. Claude went the Chrome extension route.

After testing for a couple days, some things work pretty well. Others? Wellllllll... not really.

What it means: This is encouraging from Anthropic.

Their specialty has been software engineering and coding. Right?

Now they're signaling they're not abandoning the everyday non-coding consumer. Broader audiences are in play.

Remember when OpenAI teased apps at DevDay? That's real now.

The app directory launched at chatgpt.com/apps this past week. Eight hundred plus million users can now search, discover, and install apps directly inside conversations.

Bigger sign yet? Developer submissions are now open. Previously it was controlled rollout with trusted partners only.

GitHub, Replit, Gmail, Lovable, MailChimp, Stripe. All just dropped.

Some restrictions though. Physical goods only for purchases. No digital goods, subscriptions, or in-app services yet. And everything has to be teen-appropriate.

Apps approved as soon as early 2026.

What it means: Once installed, ChatGPT will sometimes just assume you want to use a certain app based on what you say.

Pretty big deal for the ecosystem.

Google launched Gemini 3 Flash and made it the default model in the Gemini app.

Even. For. Free. Users.

Also default in AI mode in search. For free.

Sheesh. Here's where it gets wild though. Gemini 3 Flash outperforms Gemini 3 Pro on some key benchmarks, including SWE-bench Verified.

The smaller model beating the bigger model? Yup.

Google staffers said it's not a distilled version. They applied new reinforcement learning techniques that actually improved performance.

What it means: Gemini 3 Flash is a top five model on artificial analysis.

The small, cheap, fast version is extremely powerful.

Eventually Google may default search to AI mode. That's potentially bringing one of the world's most capable models to billions of users worldwide.

The US Department of Energy announced agreements with twenty-four tech organizations for the Genesis mission this past week.

AWS. Google. Microsoft. Nvidia. Intel. IBM. AMD. OpenAI. Anthropic. Oracle. HPE. Dell. Cerebras. Groq. CoreWeave. Palantir.

Pretty much every big name in tech or AI. All in.

This follows President Trump's executive order on American AI leadership. Products developed for Genesis will be architecture-agnostic.

The amount of data available here? Amounts not generally seen for frontier models.

What it means: Sovereign AI is going to be at the top of the list going into 2026.

Even if you're not an international company, the AI race is impacting things at home you may not be aware of.

Your 401k. National security. Energy. All connected.

Google snuck in a bunch of updates right when everyone was logging off for the holidays. Classic.

NotebookLM now runs on Gemini 3. Previously it was Gemini 2.5.

But that's not all. The Gemini and NotebookLM integration is now live for many paid users. Click the plus button in Gemini and boom. New NotebookLM tab.

Data tables are now available too. Synthesize facts across sources, find hidden insights, export directly to Google Sheets.

For real. One click.

What it means: NotebookLM's ability to stay grounded is the most important thing when working with large language models.

Taking unstructured data and structuring it into tables is huge.

The consulting industry should definitely be paying attention to this.

There's a new AI image model king in town. And it's not named Nano Banana.

OpenAI released GPT Image 1.5 this past week. Images up to four times faster. Preserves fine details and user intent more reliably.

Early benchmarks show it tops LM Arena's text-image leaderboard. Displaced Nano Banana Pro from first place.

Real world results are mixed though.

Nah, on the same prompts, Nano Banana Pro creates something that looks more real. Like a high-end smartphone photo. GPT Image 1.5 looks more polished. Kinda hyper-realistic. Skin tones a little glossy.

GPT Image 1.5 is probably better at iterating and understanding nuanced context. Nano Banana still wins on infographics and complex text.

What it means: It depends on what you're going for.

One-shot photos of people or venues? Nano Banana Pro still takes it.

But hey. At least it's a competition. Which means in the end, everyone wins.